Research map / Video / Video segmentation and tracking
Follow the Rules: Reasoning for Video Anomaly Detection with Large Language Models
Yuchen Yang, Kwonjoon Lee, Behzad Dariush, Yinzhi Cao, Shao-Yuan Lo
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use rule-based induction/deduction instead of asking the LLM directly | Mistral-7B on ShT · ShanghaiTech | accuracy (%): 52.1 -> 81.8 | Improved |
| add rule induction | LLM direct baseline · ShanghaiTech | recall (%): 6.2 -> 64.3 | Improved |
| add Normal and Anomaly rule contrast (ablation) | AnomalyRuler-base · ShanghaiTech | AUC (points): -18.8 AUC when removed | Improved |
| add rule aggregation (ablation) | AnomalyRuler-base · ShanghaiTech | AUC (points): -15.8 AUC when removed | Improved |
| add perception smoothing (ablation) | AnomalyRuler-base · ShanghaiTech | AUC (points): -3.3 AUC when removed | Improved |
| add Human and Environment separation (ablation) | AnomalyRuler-base · ShanghaiTech | AUC (points): -2.4 AUC when removed | Improved |
| add Abstract and Concrete strategy (ablation) | AnomalyRuler-base · ShanghaiTech | AUC (points): -0.9 AUC when removed | No clear effect |
| add robust reasoning recheck module (ablation) | AnomalyRuler-base · ShanghaiTech | AUC (%): 84.6 -> 85.2 | Improved |
| increase reference frames per batch m with n=20 (ablation) | rule aggregation · ShanghaiTech | AUC (%): 85.9 -> 72.2 | Hurt |
| increase EMA alpha from 0.09 to 0.33 (ablation) | perception smoothing · ShanghaiTech | AUC | Improved |
| use rule-based reasoning on cross-domain transfer | AnomalyRuler-base · Cross-domain Ped2/Ave/ShT/UB | AUC (%): Ped2 97.4 vs 95.7; Ave 81.6 vs 82.2 | Mixed |
Closest papers by meaning
- Vad-R1: Towards Video Anomaly Reasoning via Perception-to-Cognition Chain-of-Thought (NeurIPS 2025 · similarity 0.93)
- Harnessing Large Language Models for Training-free Video Anomaly Detection (CVPR 2024 · similarity 0.92)
- Open-Vocabulary Video Anomaly Detection (CVPR 2024 · similarity 0.92)
- VERA: Explainable Video Anomaly Detection via Verbalized Learning of Vision-Language Models (CVPR 2025 · similarity 0.90)
- MoniTor: Exploiting Large Language Models with Instruction for Online Video Anomaly Detection (NeurIPS 2025 · similarity 0.90)
- VADTree: Explainable Training-Free Video Anomaly Detection via Hierarchical Granularity-Aware Tree (NeurIPS 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
